spi: stm32_qspi: use QSPI bus as 8 lines communication channel

Merge series from patrice.chotard@foss.st.com <patrice.chotard@foss.st.com>:

The goal of this series is to allow to use QSPI bus as a 8 lines communication
channel for specific purpose.

The QSPI block offers the possibility to communicate with 2 flashes in
parrallel using the dual flash mode, 8 data lines are then used.
Usage of cs-gpios populated and spi-tx-bus-width / spi-rx-bus-width both set to 8,
is needed to enable dual flash mode.

The addition of the legacy transfer_one_message() spi callback is also needed
as currently the stm32-qspi driver only supports spi_controller_mem_ops API.

drivers/spi/spi-stm32-qspi.c

@@ -15,6 +15,7 @@
 #include <linux/mutex.h>
 #include <linux/of.h>
 #include <linux/of_device.h>
+#include <linux/of_gpio.h>
 #include <linux/pinctrl/consumer.h>
 #include <linux/pm_runtime.h>
 #include <linux/platform_device.h>
@@ -355,10 +356,10 @@ static int stm32_qspi_get_mode(u8 buswidth)
 	return buswidth;
 }
 
-static int stm32_qspi_send(struct spi_mem *mem, const struct spi_mem_op *op)
+static int stm32_qspi_send(struct spi_device *spi, const struct spi_mem_op *op)
 {
-	struct stm32_qspi *qspi = spi_controller_get_devdata(mem->spi->master);
-	struct stm32_qspi_flash *flash = &qspi->flash[mem->spi->chip_select];
+	struct stm32_qspi *qspi = spi_controller_get_devdata(spi->master);
+	struct stm32_qspi_flash *flash = &qspi->flash[spi->chip_select];
 	u32 ccr, cr;
 	int timeout, err = 0, err_poll_status = 0;
 
@@ -465,7 +466,7 @@ static int stm32_qspi_poll_status(struct spi_mem *mem, const struct spi_mem_op *
 	qspi->fmode = CCR_FMODE_APM;
 	qspi->status_timeout = timeout_ms;
 
-	ret = stm32_qspi_send(mem, op);
+	ret = stm32_qspi_send(mem->spi, op);
 	mutex_unlock(&qspi->lock);
 
 	pm_runtime_mark_last_busy(qspi->dev);
@@ -489,7 +490,7 @@ static int stm32_qspi_exec_op(struct spi_mem *mem, const struct spi_mem_op *op)
 	else
 		qspi->fmode = CCR_FMODE_INDW;
 
-	ret = stm32_qspi_send(mem, op);
+	ret = stm32_qspi_send(mem->spi, op);
 	mutex_unlock(&qspi->lock);
 
 	pm_runtime_mark_last_busy(qspi->dev);
@@ -545,7 +546,7 @@ static ssize_t stm32_qspi_dirmap_read(struct spi_mem_dirmap_desc *desc,
 	else
 		qspi->fmode = CCR_FMODE_INDR;
 
-	ret = stm32_qspi_send(desc->mem, &op);
+	ret = stm32_qspi_send(desc->mem->spi, &op);
 	mutex_unlock(&qspi->lock);
 
 	pm_runtime_mark_last_busy(qspi->dev);
@@ -554,12 +555,87 @@ static ssize_t stm32_qspi_dirmap_read(struct spi_mem_dirmap_desc *desc,
 	return ret ?: len;
 }
 
+static int stm32_qspi_transfer_one_message(struct spi_controller *ctrl,
+					   struct spi_message *msg)
+{
+	struct stm32_qspi *qspi = spi_controller_get_devdata(ctrl);
+	struct spi_transfer *transfer;
+	struct spi_device *spi = msg->spi;
+	struct spi_mem_op op;
+	int ret;
+
+	if (!spi->cs_gpiod)
+		return -EOPNOTSUPP;
+
+	mutex_lock(&qspi->lock);
+
+	gpiod_set_value_cansleep(spi->cs_gpiod, true);
+
+	list_for_each_entry(transfer, &msg->transfers, transfer_list) {
+		u8 dummy_bytes = 0;
+
+		memset(&op, 0, sizeof(op));
+
+		dev_dbg(qspi->dev, "tx_buf:%p tx_nbits:%d rx_buf:%p rx_nbits:%d len:%d dummy_data:%d\n",
+			transfer->tx_buf, transfer->tx_nbits,
+			transfer->rx_buf, transfer->rx_nbits,
+			transfer->len, transfer->dummy_data);
+
+		/*
+		 * QSPI hardware supports dummy bytes transfer.
+		 * If current transfer is dummy byte, merge it with the next
+		 * transfer in order to take into account QSPI block constraint
+		 */
+		if (transfer->dummy_data) {
+			op.dummy.buswidth = transfer->tx_nbits;
+			op.dummy.nbytes = transfer->len;
+			dummy_bytes = transfer->len;
+
+			/* if happens, means that message is not correctly built */
+			if (list_is_last(&transfer->transfer_list, &msg->transfers))
+				goto end_of_transfer;
+
+			transfer = list_next_entry(transfer, transfer_list);
+		}
+
+		op.data.nbytes = transfer->len;
+
+		if (transfer->rx_buf) {
+			qspi->fmode = CCR_FMODE_INDR;
+			op.data.buswidth = transfer->rx_nbits;
+			op.data.dir = SPI_MEM_DATA_IN;
+			op.data.buf.in = transfer->rx_buf;
+		} else {
+			qspi->fmode = CCR_FMODE_INDW;
+			op.data.buswidth = transfer->tx_nbits;
+			op.data.dir = SPI_MEM_DATA_OUT;
+			op.data.buf.out = transfer->tx_buf;
+		}
+
+		ret = stm32_qspi_send(spi, &op);
+		if (ret)
+			goto end_of_transfer;
+
+		msg->actual_length += transfer->len + dummy_bytes;
+	}
+
+end_of_transfer:
+	gpiod_set_value_cansleep(spi->cs_gpiod, false);
+
+	mutex_unlock(&qspi->lock);
+
+	msg->status = ret;
+	spi_finalize_current_message(ctrl);
+
+	return ret;
+}
+
 static int stm32_qspi_setup(struct spi_device *spi)
 {
 	struct spi_controller *ctrl = spi->master;
 	struct stm32_qspi *qspi = spi_controller_get_devdata(ctrl);
 	struct stm32_qspi_flash *flash;
-	u32 presc;
+	u32 presc, mode;
 	int ret;
 
 	if (ctrl->busy)
@@ -568,6 +644,16 @@ static int stm32_qspi_setup(struct spi_device *spi)
 	if (!spi->max_speed_hz)
 		return -EINVAL;
 
+	mode = spi->mode & (SPI_TX_OCTAL | SPI_RX_OCTAL);
+	if ((mode == SPI_TX_OCTAL || mode == SPI_RX_OCTAL) ||
+	    ((mode == (SPI_TX_OCTAL | SPI_RX_OCTAL)) &&
+	    of_gpio_named_count(qspi->dev->of_node, "cs-gpios") == -ENOENT)) {
+		dev_err(qspi->dev, "spi-rx-bus-width\\/spi-tx-bus-width\\/cs-gpios\n");
+		dev_err(qspi->dev, "configuration not supported\n");
+
+		return -EINVAL;
+	}
+
 	ret = pm_runtime_resume_and_get(qspi->dev);
 	if (ret < 0)
 		return ret;
@@ -580,6 +666,17 @@ static int stm32_qspi_setup(struct spi_device *spi)
 
 	mutex_lock(&qspi->lock);
 	qspi->cr_reg = CR_APMS | 3 << CR_FTHRES_SHIFT | CR_SSHIFT | CR_EN;
+
+	/*
+	 * Dual flash mode is only enable in case SPI_TX_OCTAL and SPI_TX_OCTAL
+	 * are both set in spi->mode and "cs-gpios" properties is found in DT
+	 */
+	if (((spi->mode & (SPI_TX_OCTAL | SPI_RX_OCTAL)) == (SPI_TX_OCTAL | SPI_RX_OCTAL)) &&
+	    of_gpio_named_count(qspi->dev->of_node, "cs-gpios")) {
+		qspi->cr_reg |= CR_DFM;
+		dev_dbg(qspi->dev, "Dual flash mode enable");
+	}
+
 	writel_relaxed(qspi->cr_reg, qspi->io_base + QSPI_CR);
 
 	/* set dcr fsize to max address */
@@ -741,11 +838,14 @@ static int stm32_qspi_probe(struct platform_device *pdev)
 
 	mutex_init(&qspi->lock);
 
-	ctrl->mode_bits = SPI_RX_DUAL | SPI_RX_QUAD
-		| SPI_TX_DUAL | SPI_TX_QUAD;
+	ctrl->mode_bits = SPI_RX_DUAL | SPI_RX_QUAD | SPI_TX_OCTAL
+		| SPI_TX_DUAL | SPI_TX_QUAD | SPI_RX_OCTAL;
 	ctrl->setup = stm32_qspi_setup;
 	ctrl->bus_num = -1;
 	ctrl->mem_ops = &stm32_qspi_mem_ops;
+	ctrl->use_gpio_descriptors = true;
+	ctrl->transfer_one_message = stm32_qspi_transfer_one_message;
+	ctrl->auto_runtime_pm = true;
 	ctrl->num_chipselect = STM32_QSPI_MAX_NORCHIP;
 	ctrl->dev.of_node = dev->of_node;